Residual bounds for unitarily invariant norms on clustered eigenvalues (Q676021): Difference between revisions

Let the \(n\times n\) Hermitian matrix \(A\) have the eigenvalues \(\lambda_1,\lambda_2,\dots,\lambda_n\), let the \(k\times k\) Hermitian matrix \(H\) have the eigenvalues \(\mu_1,\mu_2,\dots,\mu_k\), and let \(Q\) be an \(n\times k\) matrix having full column rank, so \(1\leq k\leq n\). In this short note, the author proves that there exist \(k\) eigenvalues \(\lambda_{i_1}\leq\lambda_{i_2}\leq \cdots\leq\lambda_{i_k}\) of \(A\) such that \[ |||\text{diag}(\mu_1- \lambda_{i_1}, \mu_2-\lambda_{i_2},\dots, \mu_k- \lambda_{i_k})|||\leq {c\over \sigma_{\min}(Q)} |||AQ- QH||| \] always holds with \(c=2\), where \(\sigma_{\min}(Q)\) is the smallest singular value of \(Q\) and \(|||\cdot |||\) denotes any unitarily invariant norm. The assumptions \(Q^*Q= I\) and \(H= Q^*AQ\) in \textit{G. W. Stewart} and \textit{J.-G. Sun's} book [Matrix perturbation theory (1990; Zbl 0706.65013)] are deleted. He also improves the result applicable to practical computation.